PyVisualFields 2.0.2__py3-none-any.whl → 2.0.3__py3-none-any.whl

PyVisualFields/PyGlaucoMetrics.py

@@ -1,563 +1,563 @@
-
-import numpy as np
-import pandas as pd
-
-import numpy as np
-import pandas as pd
-
-
-
-def _normalise_eye(val):
-    """Return 'OD' or 'OS' regardless of input format."""
-    v = str(val).strip().lower()
-    if v in ('od', 'right', 'r', '1', 'righteye', 're'):
-        return 'OD'
-    if v in ('os', 'left', 'l', '0', 'lefteye', 'le'):
-        return 'OS'
-    return None   # unknown — caller returns Non-GL
-
-
-def _pt_cols_from_row(row):
-    """Return sorted point column names (s* or l*) from a row's index."""
-    cols = sorted(
-        [c for c in row.index if len(c) >= 2 and c[0] in ('s', 'l') and c[1:].isdigit()],
-        key=lambda x: int(x[1:])
-    )
-    return cols
-
-
-def _pt_cols_from_df(df):
-    """Return sorted point column names (s* or l*) from a dataframe."""
-    cols = sorted(
-        [c for c in df.columns if len(c) >= 2 and c[0] in ('s', 'l') and c[1:].isdigit()],
-        key=lambda x: int(x[1:])
-    )
-    return cols
-
-
-def _eye_col(row):
-    """Return the eye value from whichever column name exists (Eye/eye/EYE)."""
-    for name in ('Eye', 'eye', 'EYE'):
-        if name in row.index:
-            return row[name]
-    return None
-
-
-def _col(prefix, n, pt_cols):
-    """
-    Return the column name for point n (1-indexed) using whatever prefix
-    is present in pt_cols (s or l).
-    """
-    actual_prefix = pt_cols[0][0] if pt_cols else 's'
-    return f'{actual_prefix}{n}'
-
-
-def convertVF_to_2D(VF):
-    VF_2D = np.full((8,9), np.nan)
-
-    VF_2D[0,3:7] = VF[0:4]
-    VF_2D[1,2:8] = VF[4:10]
-    VF_2D[2,1:9] = VF[10:18]
-    VF_2D[3,0:9] = VF[18:27]
-    VF_2D[4,0:9] = VF[27:36]
-    VF_2D[5,1:9] = VF[36:44]
-    VF_2D[6,2:8] = VF[44:50]
-    VF_2D[7,3:7] = VF[50:54]
-
-    return VF_2D
-
-##########################################################
-###################################################
-
-
-# def Fn_HAP2(df_PDP):
-#     pt_cols  = _pt_cols_from_df(df_PDP)
-#     data     = df_PDP[pt_cols]
-
-#     counts_05 = (data <= 0.05).sum(axis=1)
-#     counts_01 = (data <= 0.01).sum(axis=1)
-
-#     df_PDP['HAP2_p1_clf'] = np.where(
-#         (counts_05 >= 3) & (counts_01 >= 1), 'GL', 'Non-GL')
-
-#     df_GL = df_PDP[df_PDP['HAP2_p1_clf'] == 'GL'].copy()
-
-#     # find MD column case-insensitively
-#     md_col = next((c for c in df_PDP.columns if c.lower() in ('md', 'tmd')), None)
-#     if md_col:
-#         md = df_GL[md_col]
-#         c05 = counts_05[df_GL.index]
-#         c01 = counts_01[df_GL.index]
-#         cond1 = (md > -6.0)   & c05.between(1, 12)  & c01.between(1, 4)
-#         cond2 = (md.between(-12.0, -6.01)) | (c05.between(13, 26) & c01.between(5, 13))
-#         cond3 = (md < -12.0)  | (c05 >= 27)
-#         df_GL['HAP2_p2_clf'] = np.select(
-#             [cond1, cond2, cond3],
-#             ['Stage 1', 'Stage 2', 'Stage 3'],
-#             default='GL')
-#     else:
-#         df_GL['HAP2_p2_clf'] = 'GL'
-
-#     df_PDP.loc[df_GL.index, 'HAP2_p2_clf'] = df_GL['HAP2_p2_clf']
-#     df_PDP['HAP2_p2_clf'] = df_PDP['HAP2_p2_clf'].fillna('Non-GL')
-#     return df_PDP
-
-
-
-
-def has_pd_cluster_3_with_1_at_1pct(pd_values):
-    """
-    Criterion:
-    cluster of 3 points at P < 5%,
-    with at least 1 of those points at P < 1%.
-    """
-    pd_grid = convertVF_to_2D(pd_values)
-
-    mask_5 = pd_grid < 0.05
-    mask_1 = pd_grid < 0.01
-
-    mask_5 = np.nan_to_num(mask_5, nan=False).astype(bool)
-    mask_1 = np.nan_to_num(mask_1, nan=False).astype(bool)
-
-    visited = np.zeros(mask_5.shape, dtype=bool)
-
-    neighbors = [(-1,-1), (-1,0), (-1,1),
-                 (0,-1),           (0,1),
-                 (1,-1),  (1,0),   (1,1)]
-
-    for r in range(mask_5.shape[0]):
-        for c in range(mask_5.shape[1]):
-            if not mask_5[r, c] or visited[r, c]:
-                continue
-
-            stack = [(r, c)]
-            visited[r, c] = True
-            component = []
-
-            while stack:
-                rr, cc = stack.pop()
-                component.append((rr, cc))
-
-                for dr, dc in neighbors:
-                    nr, nc = rr + dr, cc + dc
-                    if (
-                        0 <= nr < mask_5.shape[0]
-                        and 0 <= nc < mask_5.shape[1]
-                        and mask_5[nr, nc]
-                        and not visited[nr, nc]
-                    ):
-                        visited[nr, nc] = True
-                        stack.append((nr, nc))
-
-            if len(component) >= 3:
-                has_1pct_point = any(mask_1[rr, cc] for rr, cc in component)
-                if has_1pct_point:
-                    return True
-
-    return False
-
-
-def _is_ght_outside_normal_limits(x):
-    if pd.isna(x):
-        return False
-
-    x = str(x).strip().lower()
-
-    return (
-        "outside normal limits" in x
-        or "outside" in x
-        or x in {"onl", "abnormal"}
-    )
-
-
-def _is_psd_p_less_5(row):
-    """
-    Handles possible column names:
-    PSD_P, PSD_p, PSDp, PSD_probability, PSD
-    If only PSD is present as a p-value, it checks PSD < 0.05.
-    """
-    possible_cols = [
-        "PSD_P", "PSD_p", "PSDp", "psd_p",
-        "PSD_probability", "psd_probability",
-        "PSD", "psd"
-    ]
-
-    for col in possible_cols:
-        if col in row.index:
-            val = pd.to_numeric(row[col], errors="coerce")
-            if pd.notna(val) and val < 0.05:
-                return True
-
-    return False
-
-
-def HAP2_clf(row):
-    """
-    Criteria:
-    1. GHT outside normal limits
-    OR
-    2. cluster of 3 PD probability points at P < 5%,
-       with at least 1 at P < 1%
-    OR
-    3. PSD at P < 5%
-    """
-    pt_cols = _pt_cols_from_df(pd.DataFrame([row]))
-    pd_values = pd.to_numeric(row[pt_cols], errors="coerce").values
-
-    # Criterion 1: GHT outside normal limits
-    for ght_col in ["GHT", "ght"]:
-        if ght_col in row.index and _is_ght_outside_normal_limits(row[ght_col]):
-            return "GL"
-
-    # Criterion 2: PD cluster
-    if has_pd_cluster_3_with_1_at_1pct(pd_values):
-        return "GL"
-
-    # Criterion 3: PSD P < 5%
-    if _is_psd_p_less_5(row):
-        return "GL"
-
-    return "Non-GL"
-
-
-def Fn_HAP2(df_PDP):
-    out = df_PDP.copy()
-    out["HAP2_clf"] = out.apply(HAP2_clf, axis=1)
-    return out
-
-
-
-########################################################
-########################################################
-
-# def check_gl_condition(row, threshold=0.01, consecutive_reductions=2):
-#     numeric_values = pd.to_numeric(row.values, errors='coerce')
-#     below = (numeric_values <= threshold).astype(float)
-#     below = np.nan_to_num(below, nan=0.0)
-#     counts = np.convolve(below, np.ones(consecutive_reductions), mode='valid')
-#     return np.any(counts >= consecutive_reductions)
-
-
-# def Fn_UKGTS(df_TDP, threshold=0.01, consecutive_reductions=2):
-#     pt_cols = _pt_cols_from_df(df_TDP)
-#     df_TDP['UKGTS_clf'] = np.where(
-#         df_TDP[pt_cols].apply(
-#             check_gl_condition, axis=1,
-#             threshold=threshold,
-#             consecutive_reductions=consecutive_reductions),
-#         'GL', 'Non-GL')
-#     return df_TDP
-
-
-
-def has_cluster(mask, min_size=2, connectivity=8):
-    mask = np.nan_to_num(mask, nan=False).astype(bool)
-    visited = np.zeros(mask.shape, dtype=bool)
-
-    if connectivity == 8:
-        neighbors = [(-1,-1), (-1,0), (-1,1),
-                     (0,-1),           (0,1),
-                     (1,-1),  (1,0),   (1,1)]
-    else:
-        neighbors = [(-1,0), (1,0), (0,-1), (0,1)]
-
-    for r in range(mask.shape[0]):
-        for c in range(mask.shape[1]):
-            if not mask[r, c] or visited[r, c]:
-                continue
-
-            stack = [(r, c)]
-            visited[r, c] = True
-            size = 0
-
-            while stack:
-                rr, cc = stack.pop()
-                size += 1
-
-                for dr, dc in neighbors:
-                    nr, nc = rr + dr, cc + dc
-                    if (
-                        0 <= nr < mask.shape[0]
-                        and 0 <= nc < mask.shape[1]
-                        and mask[nr, nc]
-                        and not visited[nr, nc]
-                    ):
-                        visited[nr, nc] = True
-                        stack.append((nr, nc))
-
-            if size >= min_size:
-                return True
-
-    return False
-
-def has_nasal_step_10db(td_grid, nasal_cols=(0, 1, 2, 3, 4)):
-    """
-    Assumes right eye.
-    Nasal field is assumed to be the left side of the VF grid.
-    Checks adjacent superior/inferior pairs across the horizontal midline.
-    """
-    diffs = []
-
-    for c in nasal_cols:
-        sup = td_grid[3, c]
-        inf = td_grid[4, c]
-
-        if not np.isnan(sup) and not np.isnan(inf):
-            diffs.append(abs(sup - inf) >= 10)
-        else:
-            diffs.append(False)
-
-    # cluster of 2 adjacent nasal step points
-    for i in range(len(diffs) - 1):
-        if diffs[i] and diffs[i + 1]:
-            return True
-
-    return False
-
-
-########################################################
-########################################################
-
-def UKGTS_clf(row_prob, row_db):
-    prob_values = pd.to_numeric(row_prob, errors="coerce").values
-    db_values = pd.to_numeric(row_db, errors="coerce").values
-
-    prob_grid = convertVF_to_2D(prob_values)
-    db_grid = convertVF_to_2D(db_values)
-
-    # Criterion 1: cluster of 2 points at P < 1%
-    if has_cluster(prob_grid < 0.01, min_size=2):
-        return "GL"
-
-    # Criterion 2: cluster of 3 points at P < 5%
-    if has_cluster(prob_grid < 0.05, min_size=3):
-        return "GL"
-
-    # Criterion 3: cluster of 2 points with 10 dB difference across nasal horizontal midline
-    if has_nasal_step_10db(db_grid):
-        return "GL"
-
-    return "Non-GL"
-
-def Fn_UKGTS(df_TDP, df_TD):
-    prob_cols = _pt_cols_from_df(df_TDP)
-    db_cols = _pt_cols_from_df(df_TD)
-
-    if len(prob_cols) != 54 or len(db_cols) != 54:
-        raise ValueError("Expected 54 VF point columns in both probability and dB dataframes.")
-
-    out = df_TDP.copy()
-
-    out["UKGTS_clf"] = [
-        UKGTS_clf(df_TDP.loc[i, prob_cols], df_TD.loc[i, db_cols])
-        for i in df_TDP.index
-    ]
-
-    return out
-
-
-########################################################
-########################################################
-
-
-# def LoGTS_clf(row):
-#     numeric_values = pd.to_numeric(row, errors='coerce')
-#     return 'GL' if (numeric_values < -10).sum() >= 2 else 'Non-GL'
-
-
-def Fn_LoGTS(df_TD):
-    pt_cols = _pt_cols_from_df(df_TD)
-    df = df_TD.copy()
-    df['LoGTS_clf'] = df[pt_cols].apply(LoGTS_clf, axis=1)
-    return df   
-
-
-def LoGTS_clf(row):
-    values = pd.to_numeric(row, errors='coerce').values
-    VF_2D = convertVF_to_2D(values)   # ← use the result
-
-    # criterion 1: cluster of 2 points with TD < -10 dB
-    if has_cluster(VF_2D < -10, min_size=2):
-        return "GL"
-
-    # criterion 2: cluster of 3 points with TD < -8 dB
-    if has_cluster(VF_2D < -8, min_size=3):
-        return "GL"
-
-    return "Non-GL"
-
-########################################################
-########################################################
-
-
-def Kangs_clf(row):
-    values = pd.to_numeric(row, errors='coerce').values
-    td_grid = convertVF_to_2D(values)
-
-    # criterion 1: cluster of 3 contiguous points with TD < -5 dB
-    if has_cluster(td_grid < -5, min_size=3):
-        return 'GL'
-    # criterion 2: cluster of 2 contiguous points with TD < -10 dB
-    if has_cluster(td_grid < -10, min_size=2):
-        return 'GL'
-    return 'Non-GL'
-
-
-def Fn_Kangs(df_TD):
-    pt_cols = _pt_cols_from_df(df_TD)
-    df = df_TD.copy()
-    df['Kangs_clf'] = df[pt_cols].apply(Kangs_clf, axis=1)
-    return df
-
-
-########################################################
-########################################################
-
-
-# # Sector point indices (1-based) for OD and OS — direction-corrected
-# _FOSTER_OD_SUP = [22,23,24,13,14,15,16,11,12,19,20,21,1,2,5,6,7,8,3,4,9,10]
-# _FOSTER_OD_INF = [31,32,33,39,40,41,42,28,29,30,37,38,45,46,47,48,51,52,49,50,53,54]
-# _FOSTER_OS_SUP = [22,23,24,13,14,15,16,17,18,25,26,27,3,4,8,9,10,1,2,5,6]
-# _FOSTER_OS_INF = [31,32,33,39,40,41,42,34,35,36,43,44,48,49,50,53,54,45,46,51,52]
-
-
-# def Foster_clf(row):
-#     eye = _normalise_eye(_eye_col(row))
-#     if eye is None:
-#         return 'Non-GL'
-
-#     pt_cols = _pt_cols_from_row(row)
-#     if not pt_cols:
-#         return 'Non-GL'
-#     pfx = pt_cols[0][0]   # 's' or 'l'
-
-#     if eye == 'OD':
-#         sup_cols = [f'{pfx}{n}' for n in _FOSTER_OD_SUP]
-#         inf_cols = [f'{pfx}{n}' for n in _FOSTER_OD_INF]
-#     else:
-#         sup_cols = [f'{pfx}{n}' for n in _FOSTER_OS_SUP]
-#         inf_cols = [f'{pfx}{n}' for n in _FOSTER_OS_INF]
-
-#     # keep only cols that actually exist (guards against 52-pt datasets)
-#     sup_cols = [c for c in sup_cols if c in row.index]
-#     inf_cols = [c for c in inf_cols if c in row.index]
-
-#     s_vals = pd.to_numeric(row[sup_cols], errors='coerce')
-#     i_vals = pd.to_numeric(row[inf_cols], errors='coerce')
-
-#     gl_sum_flag  = (s_vals.sum() > 0.005) and (i_vals.sum() > 0.005)
-#     gl_diff_flag = (s_vals.values - i_vals.values[:len(s_vals)]).tolist()
-#     gl_diff_flag = abs(sum(gl_diff_flag)) > 0.01
-
-#     all_pt   = pd.to_numeric(row[pt_cols], errors='coerce')
-#     count_05 = (all_pt <= 0.05).sum()
-
-#     if (gl_sum_flag or gl_diff_flag) and (count_05 >= 3):
-#         return 'GL'
-#     return 'Non-GL'
-
-
-def Foster_clf(row):
-    """
-    Foster criteria:
-
-    GHT outside normal limits
-    AND
-    cluster of 3 points at P < 5%
-    (if PD unavailable due to severe depression,
-     cluster criterion automatically satisfied)
-    """
-
-    # -----------------
-    # GHT check
-    # -----------------
-    ght_value = None
-
-    for ght_col in ["GHT", "ght"]:
-        if ght_col in row.index:
-            ght_value = row[ght_col]
-            break
-
-    if pd.isna(ght_value):
-        return "Error: missing GHT"
-
-    ght_ok = _is_ght_outside_normal_limits(ght_value)
-
-    # -----------------
-    # PD cluster check
-    # -----------------
-    pt_cols = _pt_cols_from_df(pd.DataFrame([row]))
-    pd_values = pd.to_numeric(
-        row[pt_cols],
-        errors="coerce"
-    ).values
-
-    prob_grid = convertVF_to_2D(pd_values)
-
-    cluster_ok = has_cluster(
-        prob_grid < 0.05,
-        min_size=3
-    )
-
-    # -----------------
-    # Final decision
-    # -----------------
-    return "GL" if (ght_ok and cluster_ok) else "Non-GL"
-
-
-
-def Fn_Foster(df_PDP):
-    df_P = df_PDP.copy()
-    df_P['Foster_clf'] = df_P.apply(Foster_clf, axis=1)
-    return df_P
-
-
-########################################################
-########################################################
-
-# def combine_dataframes(df_input, df1, df2, df3, df4, df5):
-#     # Combine the DataFrames along the columns axis
-#     result_combined = pd.concat([df1, df2], axis=1)
-    
-#     # Reset indices to ensure proper alignment
-#     m1_reset = df_input.reset_index(drop=True)
-#     m2_reset = result_combined.reset_index(drop=True)
-    
-#     # Add the specified columns from m1_reset to the first four columns of m2_reset
-#     m2_reset.insert(0, 'ID', m1_reset['id'])
-#     m2_reset.insert(1, 'Eye', m1_reset['eye'])
-#     m2_reset.insert(2, 'Age', m1_reset['age'])
-#     m2_reset.insert(3, 'Date', m1_reset['date'])  # Add the Date column from df_input
-    
-#     result_combined = m2_reset
-    
-#     # Concatenate df3
-#     result_combined = pd.concat([result_combined, df3.reset_index(drop=True)], axis=1)
-    
-#     # Concatenate df4
-#     result_combined = pd.concat([result_combined, df4.reset_index(drop=True)], axis=1)
-    
-#     # Concatenate df5
-#     result_combined = pd.concat([result_combined, df5.reset_index(drop=True)], axis=1)
-    
-#     # Return the combined DataFrame
-#     return result_combined
-
-# def Fn_ensemble_decision(df_input, df1, df2, df3, df4, df5, weights):
-#     # Call the combine_dataframes function with your DataFrames
-#     result_combined = combine_dataframes(df_input, df1, df2, df3, df4, df5)
-
-#     # Map categorical values to numeric equivalents for all columns except the first four
-#     for column in result_combined.columns[4:]:
-#         result_combined[column] = result_combined[column].map({'GL': 1, 'Non-GL': 0})
-
-#     # Convert data type to string to ensure consistency
-#     result_combined_en = result_combined.astype(str)
-
-#     # Calculate weighted average of predictions
-#     predictions = result_combined_en.iloc[:, 4:].apply(pd.to_numeric, errors='coerce').fillna(0)
-#     result_combined['Ensemble'] = np.average(predictions, axis=1, weights=weights)
-    
-#     return result_combined
-
-
+
+import numpy as np
+import pandas as pd
+
+import numpy as np
+import pandas as pd
+
+
+
+def _normalise_eye(val):
+    """Return 'OD' or 'OS' regardless of input format."""
+    v = str(val).strip().lower()
+    if v in ('od', 'right', 'r', '1', 'righteye', 're'):
+        return 'OD'
+    if v in ('os', 'left', 'l', '0', 'lefteye', 'le'):
+        return 'OS'
+    return None   # unknown — caller returns Non-GL
+
+
+def _pt_cols_from_row(row):
+    """Return sorted point column names (s* or l*) from a row's index."""
+    cols = sorted(
+        [c for c in row.index if len(c) >= 2 and c[0] in ('s', 'l') and c[1:].isdigit()],
+        key=lambda x: int(x[1:])
+    )
+    return cols
+
+
+def _pt_cols_from_df(df):
+    """Return sorted point column names (s* or l*) from a dataframe."""
+    cols = sorted(
+        [c for c in df.columns if len(c) >= 2 and c[0] in ('s', 'l') and c[1:].isdigit()],
+        key=lambda x: int(x[1:])
+    )
+    return cols
+
+
+def _eye_col(row):
+    """Return the eye value from whichever column name exists (Eye/eye/EYE)."""
+    for name in ('Eye', 'eye', 'EYE'):
+        if name in row.index:
+            return row[name]
+    return None
+
+
+def _col(prefix, n, pt_cols):
+    """
+    Return the column name for point n (1-indexed) using whatever prefix
+    is present in pt_cols (s or l).
+    """
+    actual_prefix = pt_cols[0][0] if pt_cols else 's'
+    return f'{actual_prefix}{n}'
+
+
+def convertVF_to_2D(VF):
+    VF_2D = np.full((8,9), np.nan)
+
+    VF_2D[0,3:7] = VF[0:4]
+    VF_2D[1,2:8] = VF[4:10]
+    VF_2D[2,1:9] = VF[10:18]
+    VF_2D[3,0:9] = VF[18:27]
+    VF_2D[4,0:9] = VF[27:36]
+    VF_2D[5,1:9] = VF[36:44]
+    VF_2D[6,2:8] = VF[44:50]
+    VF_2D[7,3:7] = VF[50:54]
+
+    return VF_2D
+
+##########################################################
+###################################################
+
+
+# def Fn_HAP2(df_PDP):
+#     pt_cols  = _pt_cols_from_df(df_PDP)
+#     data     = df_PDP[pt_cols]
+
+#     counts_05 = (data <= 0.05).sum(axis=1)
+#     counts_01 = (data <= 0.01).sum(axis=1)
+
+#     df_PDP['HAP2_p1_clf'] = np.where(
+#         (counts_05 >= 3) & (counts_01 >= 1), 'GL', 'Non-GL')
+
+#     df_GL = df_PDP[df_PDP['HAP2_p1_clf'] == 'GL'].copy()
+
+#     # find MD column case-insensitively
+#     md_col = next((c for c in df_PDP.columns if c.lower() in ('md', 'tmd')), None)
+#     if md_col:
+#         md = df_GL[md_col]
+#         c05 = counts_05[df_GL.index]
+#         c01 = counts_01[df_GL.index]
+#         cond1 = (md > -6.0)   & c05.between(1, 12)  & c01.between(1, 4)
+#         cond2 = (md.between(-12.0, -6.01)) | (c05.between(13, 26) & c01.between(5, 13))
+#         cond3 = (md < -12.0)  | (c05 >= 27)
+#         df_GL['HAP2_p2_clf'] = np.select(
+#             [cond1, cond2, cond3],
+#             ['Stage 1', 'Stage 2', 'Stage 3'],
+#             default='GL')
+#     else:
+#         df_GL['HAP2_p2_clf'] = 'GL'
+
+#     df_PDP.loc[df_GL.index, 'HAP2_p2_clf'] = df_GL['HAP2_p2_clf']
+#     df_PDP['HAP2_p2_clf'] = df_PDP['HAP2_p2_clf'].fillna('Non-GL')
+#     return df_PDP
+
+
+
+
+def has_pd_cluster_3_with_1_at_1pct(pd_values):
+    """
+    Criterion:
+    cluster of 3 points at P < 5%,
+    with at least 1 of those points at P < 1%.
+    """
+    pd_grid = convertVF_to_2D(pd_values)
+
+    mask_5 = pd_grid < 0.05
+    mask_1 = pd_grid < 0.01
+
+    mask_5 = np.nan_to_num(mask_5, nan=False).astype(bool)
+    mask_1 = np.nan_to_num(mask_1, nan=False).astype(bool)
+
+    visited = np.zeros(mask_5.shape, dtype=bool)
+
+    neighbors = [(-1,-1), (-1,0), (-1,1),
+                 (0,-1),           (0,1),
+                 (1,-1),  (1,0),   (1,1)]
+
+    for r in range(mask_5.shape[0]):
+        for c in range(mask_5.shape[1]):
+            if not mask_5[r, c] or visited[r, c]:
+                continue
+
+            stack = [(r, c)]
+            visited[r, c] = True
+            component = []
+
+            while stack:
+                rr, cc = stack.pop()
+                component.append((rr, cc))
+
+                for dr, dc in neighbors:
+                    nr, nc = rr + dr, cc + dc
+                    if (
+                        0 <= nr < mask_5.shape[0]
+                        and 0 <= nc < mask_5.shape[1]
+                        and mask_5[nr, nc]
+                        and not visited[nr, nc]
+                    ):
+                        visited[nr, nc] = True
+                        stack.append((nr, nc))
+
+            if len(component) >= 3:
+                has_1pct_point = any(mask_1[rr, cc] for rr, cc in component)
+                if has_1pct_point:
+                    return True
+
+    return False
+
+
+def _is_ght_outside_normal_limits(x):
+    if pd.isna(x):
+        return False
+
+    x = str(x).strip().lower()
+
+    return (
+        "outside normal limits" in x
+        or "outside" in x
+        or x in {"onl", "abnormal"}
+    )
+
+
+def _is_psd_p_less_5(row):
+    """
+    Handles possible column names:
+    PSD_P, PSD_p, PSDp, PSD_probability, PSD
+    If only PSD is present as a p-value, it checks PSD < 0.05.
+    """
+    possible_cols = [
+        "PSD_P", "PSD_p", "PSDp", "psd_p",
+        "PSD_probability", "psd_probability",
+        "PSD", "psd"
+    ]
+
+    for col in possible_cols:
+        if col in row.index:
+            val = pd.to_numeric(row[col], errors="coerce")
+            if pd.notna(val) and val < 0.05:
+                return True
+
+    return False
+
+
+def HAP2_clf(row):
+    """
+    Criteria:
+    1. GHT outside normal limits
+    OR
+    2. cluster of 3 PD probability points at P < 5%,
+       with at least 1 at P < 1%
+    OR
+    3. PSD at P < 5%
+    """
+    pt_cols = _pt_cols_from_df(pd.DataFrame([row]))
+    pd_values = pd.to_numeric(row[pt_cols], errors="coerce").values
+
+    # Criterion 1: GHT outside normal limits
+    for ght_col in ["GHT", "ght"]:
+        if ght_col in row.index and _is_ght_outside_normal_limits(row[ght_col]):
+            return "GL"
+
+    # Criterion 2: PD cluster
+    if has_pd_cluster_3_with_1_at_1pct(pd_values):
+        return "GL"
+
+    # Criterion 3: PSD P < 5%
+    if _is_psd_p_less_5(row):
+        return "GL"
+
+    return "Non-GL"
+
+
+def Fn_HAP2(df_PDP):
+    out = df_PDP.copy()
+    out["HAP2_clf"] = out.apply(HAP2_clf, axis=1)
+    return out
+
+
+
+########################################################
+########################################################
+
+# def check_gl_condition(row, threshold=0.01, consecutive_reductions=2):
+#     numeric_values = pd.to_numeric(row.values, errors='coerce')
+#     below = (numeric_values <= threshold).astype(float)
+#     below = np.nan_to_num(below, nan=0.0)
+#     counts = np.convolve(below, np.ones(consecutive_reductions), mode='valid')
+#     return np.any(counts >= consecutive_reductions)
+
+
+# def Fn_UKGTS(df_TDP, threshold=0.01, consecutive_reductions=2):
+#     pt_cols = _pt_cols_from_df(df_TDP)
+#     df_TDP['UKGTS_clf'] = np.where(
+#         df_TDP[pt_cols].apply(
+#             check_gl_condition, axis=1,
+#             threshold=threshold,
+#             consecutive_reductions=consecutive_reductions),
+#         'GL', 'Non-GL')
+#     return df_TDP
+
+
+
+def has_cluster(mask, min_size=2, connectivity=8):
+    mask = np.nan_to_num(mask, nan=False).astype(bool)
+    visited = np.zeros(mask.shape, dtype=bool)
+
+    if connectivity == 8:
+        neighbors = [(-1,-1), (-1,0), (-1,1),
+                     (0,-1),           (0,1),
+                     (1,-1),  (1,0),   (1,1)]
+    else:
+        neighbors = [(-1,0), (1,0), (0,-1), (0,1)]
+
+    for r in range(mask.shape[0]):
+        for c in range(mask.shape[1]):
+            if not mask[r, c] or visited[r, c]:
+                continue
+
+            stack = [(r, c)]
+            visited[r, c] = True
+            size = 0
+
+            while stack:
+                rr, cc = stack.pop()
+                size += 1
+
+                for dr, dc in neighbors:
+                    nr, nc = rr + dr, cc + dc
+                    if (
+                        0 <= nr < mask.shape[0]
+                        and 0 <= nc < mask.shape[1]
+                        and mask[nr, nc]
+                        and not visited[nr, nc]
+                    ):
+                        visited[nr, nc] = True
+                        stack.append((nr, nc))
+
+            if size >= min_size:
+                return True
+
+    return False
+
+def has_nasal_step_10db(td_grid, nasal_cols=(0, 1, 2, 3, 4)):
+    """
+    Assumes right eye.
+    Nasal field is assumed to be the left side of the VF grid.
+    Checks adjacent superior/inferior pairs across the horizontal midline.
+    """
+    diffs = []
+
+    for c in nasal_cols:
+        sup = td_grid[3, c]
+        inf = td_grid[4, c]
+
+        if not np.isnan(sup) and not np.isnan(inf):
+            diffs.append(abs(sup - inf) >= 10)
+        else:
+            diffs.append(False)
+
+    # cluster of 2 adjacent nasal step points
+    for i in range(len(diffs) - 1):
+        if diffs[i] and diffs[i + 1]:
+            return True
+
+    return False
+
+
+########################################################
+########################################################
+
+def UKGTS_clf(row_prob, row_db):
+    prob_values = pd.to_numeric(row_prob, errors="coerce").values
+    db_values = pd.to_numeric(row_db, errors="coerce").values
+
+    prob_grid = convertVF_to_2D(prob_values)
+    db_grid = convertVF_to_2D(db_values)
+
+    # Criterion 1: cluster of 2 points at P < 1%
+    if has_cluster(prob_grid < 0.01, min_size=2):
+        return "GL"
+
+    # Criterion 2: cluster of 3 points at P < 5%
+    if has_cluster(prob_grid < 0.05, min_size=3):
+        return "GL"
+
+    # Criterion 3: cluster of 2 points with 10 dB difference across nasal horizontal midline
+    if has_nasal_step_10db(db_grid):
+        return "GL"
+
+    return "Non-GL"
+
+def Fn_UKGTS(df_TDP, df_TD):
+    prob_cols = _pt_cols_from_df(df_TDP)
+    db_cols = _pt_cols_from_df(df_TD)
+
+    if len(prob_cols) != 54 or len(db_cols) != 54:
+        raise ValueError("Expected 54 VF point columns in both probability and dB dataframes.")
+
+    out = df_TDP.copy()
+
+    out["UKGTS_clf"] = [
+        UKGTS_clf(df_TDP.loc[i, prob_cols], df_TD.loc[i, db_cols])
+        for i in df_TDP.index
+    ]
+
+    return out
+
+
+########################################################
+########################################################
+
+
+# def LoGTS_clf(row):
+#     numeric_values = pd.to_numeric(row, errors='coerce')
+#     return 'GL' if (numeric_values < -10).sum() >= 2 else 'Non-GL'
+
+
+def Fn_LoGTS(df_TD):
+    pt_cols = _pt_cols_from_df(df_TD)
+    df = df_TD.copy()
+    df['LoGTS_clf'] = df[pt_cols].apply(LoGTS_clf, axis=1)
+    return df   
+
+
+def LoGTS_clf(row):
+    values = pd.to_numeric(row, errors='coerce').values
+    VF_2D = convertVF_to_2D(values)   # ← use the result
+
+    # criterion 1: cluster of 2 points with TD < -10 dB
+    if has_cluster(VF_2D < -10, min_size=2):
+        return "GL"
+
+    # criterion 2: cluster of 3 points with TD < -8 dB
+    if has_cluster(VF_2D < -8, min_size=3):
+        return "GL"
+
+    return "Non-GL"
+
+########################################################
+########################################################
+
+
+def Kangs_clf(row):
+    values = pd.to_numeric(row, errors='coerce').values
+    td_grid = convertVF_to_2D(values)
+
+    # criterion 1: cluster of 3 contiguous points with TD < -5 dB
+    if has_cluster(td_grid < -5, min_size=3):
+        return 'GL'
+    # criterion 2: cluster of 2 contiguous points with TD < -10 dB
+    if has_cluster(td_grid < -10, min_size=2):
+        return 'GL'
+    return 'Non-GL'
+
+
+def Fn_Kangs(df_TD):
+    pt_cols = _pt_cols_from_df(df_TD)
+    df = df_TD.copy()
+    df['Kangs_clf'] = df[pt_cols].apply(Kangs_clf, axis=1)
+    return df
+
+
+########################################################
+########################################################
+
+
+# # Sector point indices (1-based) for OD and OS — direction-corrected
+# _FOSTER_OD_SUP = [22,23,24,13,14,15,16,11,12,19,20,21,1,2,5,6,7,8,3,4,9,10]
+# _FOSTER_OD_INF = [31,32,33,39,40,41,42,28,29,30,37,38,45,46,47,48,51,52,49,50,53,54]
+# _FOSTER_OS_SUP = [22,23,24,13,14,15,16,17,18,25,26,27,3,4,8,9,10,1,2,5,6]
+# _FOSTER_OS_INF = [31,32,33,39,40,41,42,34,35,36,43,44,48,49,50,53,54,45,46,51,52]
+
+
+# def Foster_clf(row):
+#     eye = _normalise_eye(_eye_col(row))
+#     if eye is None:
+#         return 'Non-GL'
+
+#     pt_cols = _pt_cols_from_row(row)
+#     if not pt_cols:
+#         return 'Non-GL'
+#     pfx = pt_cols[0][0]   # 's' or 'l'
+
+#     if eye == 'OD':
+#         sup_cols = [f'{pfx}{n}' for n in _FOSTER_OD_SUP]
+#         inf_cols = [f'{pfx}{n}' for n in _FOSTER_OD_INF]
+#     else:
+#         sup_cols = [f'{pfx}{n}' for n in _FOSTER_OS_SUP]
+#         inf_cols = [f'{pfx}{n}' for n in _FOSTER_OS_INF]
+
+#     # keep only cols that actually exist (guards against 52-pt datasets)
+#     sup_cols = [c for c in sup_cols if c in row.index]
+#     inf_cols = [c for c in inf_cols if c in row.index]
+
+#     s_vals = pd.to_numeric(row[sup_cols], errors='coerce')
+#     i_vals = pd.to_numeric(row[inf_cols], errors='coerce')
+
+#     gl_sum_flag  = (s_vals.sum() > 0.005) and (i_vals.sum() > 0.005)
+#     gl_diff_flag = (s_vals.values - i_vals.values[:len(s_vals)]).tolist()
+#     gl_diff_flag = abs(sum(gl_diff_flag)) > 0.01
+
+#     all_pt   = pd.to_numeric(row[pt_cols], errors='coerce')
+#     count_05 = (all_pt <= 0.05).sum()
+
+#     if (gl_sum_flag or gl_diff_flag) and (count_05 >= 3):
+#         return 'GL'
+#     return 'Non-GL'
+
+
+def Foster_clf(row):
+    """
+    Foster criteria:
+
+    GHT outside normal limits
+    AND
+    cluster of 3 points at P < 5%
+    (if PD unavailable due to severe depression,
+     cluster criterion automatically satisfied)
+    """
+
+    # -----------------
+    # GHT check
+    # -----------------
+    ght_value = None
+
+    for ght_col in ["GHT", "ght"]:
+        if ght_col in row.index:
+            ght_value = row[ght_col]
+            break
+
+    if pd.isna(ght_value):
+        return "Error: missing GHT"
+
+    ght_ok = _is_ght_outside_normal_limits(ght_value)
+
+    # -----------------
+    # PD cluster check
+    # -----------------
+    pt_cols = _pt_cols_from_df(pd.DataFrame([row]))
+    pd_values = pd.to_numeric(
+        row[pt_cols],
+        errors="coerce"
+    ).values
+
+    prob_grid = convertVF_to_2D(pd_values)
+
+    cluster_ok = has_cluster(
+        prob_grid < 0.05,
+        min_size=3
+    )
+
+    # -----------------
+    # Final decision
+    # -----------------
+    return "GL" if (ght_ok and cluster_ok) else "Non-GL"
+
+
+
+def Fn_Foster(df_PDP):
+    df_P = df_PDP.copy()
+    df_P['Foster_clf'] = df_P.apply(Foster_clf, axis=1)
+    return df_P
+
+
+########################################################
+########################################################
+
+# def combine_dataframes(df_input, df1, df2, df3, df4, df5):
+#     # Combine the DataFrames along the columns axis
+#     result_combined = pd.concat([df1, df2], axis=1)
+    
+#     # Reset indices to ensure proper alignment
+#     m1_reset = df_input.reset_index(drop=True)
+#     m2_reset = result_combined.reset_index(drop=True)
+    
+#     # Add the specified columns from m1_reset to the first four columns of m2_reset
+#     m2_reset.insert(0, 'ID', m1_reset['id'])
+#     m2_reset.insert(1, 'Eye', m1_reset['eye'])
+#     m2_reset.insert(2, 'Age', m1_reset['age'])
+#     m2_reset.insert(3, 'Date', m1_reset['date'])  # Add the Date column from df_input
+    
+#     result_combined = m2_reset
+    
+#     # Concatenate df3
+#     result_combined = pd.concat([result_combined, df3.reset_index(drop=True)], axis=1)
+    
+#     # Concatenate df4
+#     result_combined = pd.concat([result_combined, df4.reset_index(drop=True)], axis=1)
+    
+#     # Concatenate df5
+#     result_combined = pd.concat([result_combined, df5.reset_index(drop=True)], axis=1)
+    
+#     # Return the combined DataFrame
+#     return result_combined
+
+# def Fn_ensemble_decision(df_input, df1, df2, df3, df4, df5, weights):
+#     # Call the combine_dataframes function with your DataFrames
+#     result_combined = combine_dataframes(df_input, df1, df2, df3, df4, df5)
+
+#     # Map categorical values to numeric equivalents for all columns except the first four
+#     for column in result_combined.columns[4:]:
+#         result_combined[column] = result_combined[column].map({'GL': 1, 'Non-GL': 0})
+
+#     # Convert data type to string to ensure consistency
+#     result_combined_en = result_combined.astype(str)
+
+#     # Calculate weighted average of predictions
+#     predictions = result_combined_en.iloc[:, 4:].apply(pd.to_numeric, errors='coerce').fillna(0)
+#     result_combined['Ensemble'] = np.average(predictions, axis=1, weights=weights)
+    
+#     return result_combined
+
+